Narrow parking space parking optimization method based on intelligent driving parking area formalized design

Abstract

The invention discloses a narrow parking space parking optimization method based on intelligent driving parking area formalized design, and relates to the technical field of automatic driving, and the method comprises the following steps: building a parking area grid map through multi-sensor fusion, detecting a narrow parking space, and generating an environment model; if the data is lost, the parking space detection fails or the type judgment is wrong, the step is executed again; dividing a parking area into three functional areas including a preparation area, an operation area and an adjustment area, defining multi-step parking stages including coarse adjustment, fine adjustment and fine adjustment, respectively matching the stages with corresponding neighborhoods, and returning to the step if no feasible path exists in subsequent path search for more than or equal to 3 times due to unreasonable area division; according to the method, a multi-stage parking framework is constructed through regional division and conquering on the basis of an intelligent driving parking domain formalized design concept, and narrow parking space path smooth planning, efficient search and dynamic obstacle avoidance are realized in combination with an improved mixed A * algorithm and a dynamic optimization strategy, so that the adaptability and reliability of automatic parking in a complex narrow environment are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving, in particular to a narrow parking space parking optimization method based on intelligent driving parking area formalization design. BACKGROUND

[0002] With the rapid iteration of intelligent driving technology, automatic parking as the core function to improve the intelligent experience of vehicles, the application scenario continues to expand, and the demand of users for parking accuracy, efficiency and safety is increasingly demanding. Under the background of the shortage of urban parking resources, the proportion of narrow parking spaces such as vertical compact parking spaces and parallel narrow lane parking spaces is rising, which has become a scene that automatic parking needs to break through. The space constraint of narrow parking space scene is strict, and the distribution of adjacent vehicles, walls and other obstacles is dense, and the operation fault tolerance is extremely low. The vehicle needs to complete accurate steering cut-in, attitude adjustment and final alignment in this scene, which not only needs to avoid collision, but also needs to ensure smooth action, which puts high requirements on path planning and control strategy.

[0003] The existing automatic parking path planning technology has obvious limitations: the discontinuity of path curvature leads to unstable control, resulting in steering wheel shaking and speed fluctuation; the search algorithm has high redundancy, which is difficult to converge to the optimal path in a limited space, and the real-time performance is insufficient; the static safety margin cannot cope with the intrusion of dynamic obstacles such as pedestrians and moving vehicles, and there is a risk of safety; the multi-stage parking path is hard to connect, which is easy to present a local optimal solution, resulting in a high failure rate. Therefore, a narrow parking space parking optimization method based on intelligent driving parking area formalization design is needed to solve the above problems. SUMMARY

[0004] The purpose of the present application is to provide a narrow parking space parking optimization method based on intelligent driving parking area formalization design to solve the problems existing in the prior art proposed in the background.

[0005] To achieve the above purpose, the present application provides the following technical scheme: A narrow parking space parking optimization method based on intelligent driving parking area formalization design, comprising the following steps: S1: Construct a parking area grid map by multi-sensor fusion, detect a narrow parking space and generate an environment model, and if data is lost, parking detection fails or type determination is wrong, re-execute this step; S2: Divide the parking area into three functional areas: preparation area, operation area and adjustment area, define multi-step parking stages of coarse adjustment, fine adjustment and fine adjustment, and each stage is matched with the corresponding neighborhood, and if the subsequent path search is continuous for no more than 3 times due to unreasonable area division, return to this step; S3: Multi-stage path search based on improved hybrid A* algorithm with bidirectional search and heuristic pruning rules, if the search result has local optimum or no feasible path due to excessive pruning, adjust the parameters and re-execute this step; S4: Smooth multi-step path with piecewise B-spline interpolation, and strengthen obstacle avoidance with adaptive safety weight, if the curvature of the smoothed path is suddenly changed or the safety distance is not up to standard, return to S3; S5: Real-time adjustment of multi-step parking path based on dynamic grid update and MPC control, if the parking is terminated due to the intrusion of obstacles, return to S3.

[0006] Preferably, in S1, when WP2≤WP≤WP1 is satisfied, it is determined as a narrow parking space, wherein WP is the parking space width, WV is the vehicle width, WP1=1.35WV, WP2=1.25WV, and the narrow parking space includes vertical, parallel, and diagonal types.

[0007] Preferably, in S2, the vehicle's region is quickly determined by grid indexing, and the parking area is divided into a preparation zone, an operation zone, and an adjustment zone. The preparation zone is used for parking space detection and initial positioning as a coarse adjustment stage, which can use rectangular neighborhood and trapezoidal neighborhood for cooperation. The operation zone is the core turning-in region, and the vehicle is moved into the parking space through steering and speed control as a fine adjustment stage, which can use trapezoidal neighborhood and ring neighborhood for cooperation. The adjustment zone completes the fine tuning and final parking as a fine tuning stage, which can use sector neighborhood and rectangular neighborhood for cooperation.

[0008] Preferably, in S3, under the region division framework, the improved hybrid A* algorithm is used to generate a feasible path by adapting to the characteristics of the region, and a bidirectional multi-stage search algorithm is used to generate a multi-step parking path: Coarse adjustment stage: In the preparation zone, use rectangular and trapezoidal neighborhood to generate a straight line path from the starting point to the entrance of the operation zone through the improved hybrid A* algorithm, and the heuristic function is ; wherein, is the total cost of node n in the coarse adjustment stage, is the actual cost from the starting point of the coarse adjustment stage to node n, w1 is the distance weight coefficient, is the Euclidean distance from the current node to the entrance of the operation zone; Fine adjustment stage: In the operation zone, use trapezoidal and ring neighborhood to start bidirectional search, and preferentially generate circular arc path to deal with small curvature turning, and the cost function is:

[0009] wherein, Total cost of node n in fine-tuning stage, Actual cost from the starting point of fine-tuning stage to node n, Actual cost from the target point to n, Path curvature, Node-to-obstacle distance, w2 and w3 are weight coefficients; Fine-tuning stage: Use sector and rectangular neighborhood in the adjustment area, and search the straight path in reverse from the target point to ensure that the angle between the vehicle body and the parking line is ≤5°.

[0010] Preferably, in S3, the heuristic pruning rule is as follows: Coarse tuning stage: Remove nodes with curvature exceeding the minimum turning curvature of the vehicle, and remove the path segment when the turning angle change rate of the last three nodes is >5° / s; Fine-tuning stage: Use dynamic threshold pruning, remove nodes with a cost exceeding 1.2 times the current optimal path when the obstacle density in the operating area is >3 / m2, and remove nodes with a cost exceeding 1.8 times the current optimal path when the obstacle density in the operating area is ≤3 / m2; At the same time, in bidirectional search, if a node is a leaf node, i.e. it has no child nodes, and its cost exceeds 1.5 times the current optimal path, then delete the node and its parent node; Fine-tuning stage: Remove nodes with a distance to the obstacle less than 0.3m, a cost exceeding 1.2 times the current optimal path, and an angle between the vehicle body and the parking line >5°.

[0011] Preferably, in S4, the segmented path smoothing and safety optimization includes: using 3rd order B-spline interpolation for the connection path between the coarse tuning stage and the fine tuning stage to ensure curvature continuity; using 2nd order B-spline interpolation for the connection path between the fine tuning stage and the fine tuning stage to reduce computational complexity; dynamically adjusting the path safety item according to the obstacle distance, when the obstacle distance The safety weight forces the path to move away from the obstacle, otherwise balances the path smoothness and safety; solve the optimization objective function by quadratic programming (QP) to force the path to be at least from the obstacle; the optimization objective function is:

[0012] wherein, is the path tracking error vector, is the control amount increment, and Q and R are weight matrices.

[0013] Preferably, in S5, the MPC control adapts to the three-stage characteristics using a differentiated strategy: Coarse tuning stage: MPC uses a shorter prediction time domain Tp1, focuses on minimizing path tracking error, and the control amount penalty coefficient 1=0.5, ensure the initial path stability; Fine tuning stage: MPC every 500ms roll optimization front wheel rotation angle, using a medium prediction horizon Tp2, improve the curvature adaptability, control amount penalty coefficient 2=1.0, balance steering flexibility and smoothness; Fine tuning stage: MPC uses a very short prediction horizon Tp3, increase the control amount penalty coefficient 3=2.0, limit the front wheel rotation angle increment, realize millimeter level position correction; When detecting obstacles invading the operating area ring neighborhood or trapezoidal neighborhood, multi-step parking retry is triggered, and according to the specific circumstances, it is returned to S3 for path planning to reach the expected place.

[0014] Preferably, in the S5, the optimization objective of the MPC algorithm is:

[0015] The constraint condition is: front wheel rotation angle , vehicle speed , wherein is the prediction error, is the steering angle increment, is the control amount penalty coefficient.

[0016] Compared with the prior art, the beneficial effects of the present application are: 1、The present application constructs a parking area grid map by multi-sensor fusion to detect narrow parking spaces and generate an environment model, divides multiple functional areas, matches multiple neighborhood corresponding coarse tuning, fine tuning and fine tuning stages, searches a path based on an improved hybrid A* algorithm, smoothes through segmented B-spline interpolation, avoids obstacles in combination with adaptive safety weight, and finally realizes real-time adjustment through dynamic grid update and MPC phased control to form a complete technical system.

[0017] 2、The present application ensures the continuity of path curvature, improves smoothness, improves algorithm and pruning to reduce redundancy, improves efficiency and real-time performance, strengthens obstacle avoidance with adaptive weight, reduces collision risk, and MPC phased control takes into account stability and fine tuning accuracy to avoid local optimum, thereby significantly improving the smoothness, safety and real-time performance of the narrow parking path, effectively dealing with dynamic obstacles, reducing the failure rate, and enhancing the adaptability of automatic parking in space constrained scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0018] Fig. 1 The present application is a basic flowchart of the method.

[0019] Fig. 2 The present application is a description of the narrow parking neighborhood in the embodiment.

[0020] Fig. 3 Figure of the narrow parking optimization method in the embodiment of the application. DETAILED DESCRIPTION

[0021] In order to make the technical means, creative features, purposes and effects of the present application easy to understand, the following will further describe the present application in combination with specific embodiments.

[0022] Please refer to Figs. 1-3 The present application provides the following technical solutions: A narrow parking space parking optimization method based on intelligent driving parking area formalization design, comprising the following steps: S1: Construct a parking area grid map through multi-sensor fusion, detect a narrow parking space and generate an environment model, and if data is lost, parking space detection fails or type determination is wrong, re-execute this step; Construct a parking area grid map through multi-sensor fusion such as laser radar and visual camera, and dynamically detect a narrow parking space based on topological structure. When WP2≤WP≤WP1, it is determined as a narrow parking space, wherein WP is the parking space width, WV is the vehicle width, WP1=1.35WV, and WP2=1.25WV. The narrow parking space includes vertical, parallel, and oblique types.

[0023] S2: Divide the parking area into three functional areas of preparation area, operation area, and adjustment area, define multi-step parking stages of coarse adjustment, fine adjustment, and fine adjustment, and each stage is matched with a corresponding neighborhood. If the subsequent path search is continuous≥3 times without a feasible path due to unreasonable area division, return to this step; Quickly judge the area where the vehicle is located through grid index, divide the parking area into preparation area, operation area, and adjustment area; the preparation area is used for completing parking space detection and initial positioning as the coarse adjustment stage, which can adopt rectangular neighborhood and trapezoidal neighborhood for cooperation; the operation area is the core turning cut-in area, which moves the vehicle into the parking space through turning and speed control as the fine adjustment stage, which can adopt trapezoidal neighborhood and ring neighborhood for cooperation; the adjustment area completes fine adjustment and final parking stability after parking as the fine adjustment stage, which can adopt sector neighborhood and rectangular neighborhood for cooperation. Three different speed thresholds are used to divide the start and end of each area.

[0024] S3: Perform multi-stage path search based on a hybrid A* algorithm improved based on bidirectional search and heuristic pruning rules. If the search result exists local optimum or there is no feasible path due to excessive pruning, adjust the parameters and re-execute this step; Under the area division framework, the improved hybrid A* algorithm is used to generate a feasible path by adapting to the area characteristics, and a bidirectional multi-stage search algorithm is used to generate a multi-step parking path: Coarse stage: using rectangular and trapezoidal neighborhood in preparation area, generating straight path from start point to operation area entrance by improving hybrid A* algorithm, heuristic function is ; wherein, is total cost of node n in coarse stage, is actual cost from start point of coarse stage to node n, w1 is distance weight coefficient, is Euclidean distance from current node to operation area entrance; Fine stage: using trapezoidal and ring neighborhood in operation area, starting bidirectional search (start point→target point+target point→start point), giving priority to generating circular arc path to deal with small curvature turning, cost function is:

[0025] wherein, is total cost of node n in fine stage, is actual cost from start point of fine stage to node n, is actual cost from target point to n, is path curvature, is distance from node to obstacle, w2, w3 are weight coefficients; Fine stage: using trapezoidal and ring neighborhood in operation area, starting bidirectional search (start point→target point+target point→start point), giving priority to generating circular arc path to deal with small curvature turning, cost function is:

[0026] Heuristic pruning rules are as follows: Coarse stage: eliminating nodes with curvature exceeding the minimum turning curvature of the vehicle, and when the turning angle change rate of the last three nodes is >5° / s, the path segment is eliminated; Fine stage: dynamic threshold pruning is adopted, when the obstacle density in operation area is >3 / m2, nodes with cost exceeding 1.2 times of the current optimal path are eliminated, when the obstacle density in operation area is ≤3 / m2, nodes with cost exceeding 1.8 times of the current optimal path are eliminated; at the same time, in bidirectional search, if a node is a leaf node, i.e. it has no child nodes, and its cost exceeds 1.5 times of the current optimal path, the node and its parent node are deleted; Fine stage: eliminating nodes with distance to obstacle less than 0.3m, cost exceeding 1.2 times of the current optimal path and angle between vehicle body and parking space boundary >5°.

[0027] S4: using piecewise B-spline interpolation to smooth the multi-step path, and strengthening obstacle avoidance by adaptive safety weight, if the curvature of the smoothed path is suddenly changed or the safety distance is not up to standard, return to S3; Piecewise path smoothing and safety optimization includes: using cubic B-spline interpolation for the connection path of coarse stage and fine stage to ensure curvature continuity (C 2The connection path between the fine adjustment stage and the fine-tuning stage adopts 2 times B-spline interpolation, and the calculation complexity is reduced; the path safety item is dynamically adjusted according to the distance of the obstacle When the safety weight is greater than 0, the path is forced to be away from the obstacle, otherwise The path smoothness and safety are balanced; the optimization objective function is solved by quadratic programming (QP), and the distance between the path and the obstacle is forced to be greater than The optimization objective function is:

[0028] Wherein, is the path tracking error vector, is the control amount increment, Q and R are weight matrices.

[0029] S5: Based on dynamic grid update and MPC control, real-time adjustment of multi-step parking path is realized, and if the parking is terminated due to the intrusion of obstacles, it returns to S3.

[0030] The MPC control adapts to the three-stage characteristics by using a differentiated strategy: Coarse adjustment stage: MPC uses a shorter prediction time domain Tp1 (such as Tp1=3-5 steps), focuses on minimizing the path tracking error, and the control amount penalty coefficient 1=0.5, to ensure the stability of the initial path; Fine adjustment stage: MPC rolls the front wheel steering angle every 500ms, uses a medium prediction time domain Tp2 (such as Tp2=5-8 steps), improves the curvature adaptability, and the control amount penalty coefficient 2=1.0, to balance the steering flexibility and smoothness; Fine-tuning stage: MPC uses an extremely short prediction time domain Tp3 (such as Tp3=2-3 steps), increases the control amount penalty coefficient 3=2.0, to limit the front wheel steering angle increment (such as Δδ≤±1°), to achieve millimeter-level position correction; When the detection of obstacles invading the operating area annulus or trapezoidal neighborhood is triggered, multi-step parking retry is triggered, and according to the specific circumstances, it returns to S3 for path planning to reach the expected location.

[0031] The optimization objective of the MPC algorithm is:

[0032] The constraint condition is that the front wheel steering angle , the vehicle speed , wherein is the prediction error, is the steering angle increment, is the control amount penalty coefficient.

[0033] Embodiment According to the above, taking a narrow vertical parking space as an example. A mathematical model is established to determine the neighborhood range that the vehicle trajectory can pass through in the coordinate system. For example, in the attribute model, the controlled vehicle is represented by a circle, the black body rectangle represents the fixed obstacles in the controlled vehicle driving environment, the black body triangle represents the moving obstacles, and T1-T6 represents the position of the controlled vehicle.

[0034] After detecting a narrow vertical parking space by multiple sensors, it directly enters the parking area division stage. Among them, the preparation area T1-T2 controls the vehicle speed to be lower than the preset first threshold (such as ≤5 km / h), and the front wheel steering angle is limited within a certain threshold (such as ±10°) to reduce the target speed with negative acceleration; the operation area T2-T5 controls the vehicle speed to be lower than the preset second threshold (such as ≤3 km / h), dynamically adjusts the acceleration according to the real-time path deviation, and the front wheel steering angle uses a certain threshold of steering angle (such as ±30° or more) according to the parking type; the adjustment area T5-T6 controls the vehicle speed to be lower than the preset third threshold (such as ≤1 km / h), and the position is fine-tuned by periodic short acceleration and deceleration, and the front wheel steering angle is limited within a smaller threshold (such as ±5°).

[0035] The action performed by the controlled vehicle in the preparation area is the coarse adjustment stage, at which time the rectangular neighborhood is used in the preparation area T1-T2, and the straight line path from the starting point to the entrance of the operation area is generated by improving the hybrid A* algorithm using forward search, and the heuristic function is , the pruning rule is to remove the nodes with curvature exceeding the minimum turning curvature of the vehicle, and when the steering angle change rate of the continuous 3 nodes is >5° / s, the path segment is removed.

[0036] The action performed by the controlled vehicle in the operation area is the fine adjustment stage, at which time the trapezoidal and ring-shaped neighborhoods are used in the operation area T2-T5, and bidirectional search is started (operation area starting point T2→adjustment area target point T5, adjustment area target point T5→operation area starting point T2), and the circular arc path is generated to cope with small curvature turning. The cost function is:

[0037] The pruning rule adopts dynamic threshold pruning, when the obstacle density of the operation area is >3 / ㎡, the nodes with cost exceeding 1.2 times of the current optimal path are removed, when the obstacle density of the operation area is ≤3 / ㎡, the nodes with cost exceeding 1.8 times of the current optimal path are removed; at the same time, in the bidirectional search, if a node is a leaf node (no child node) and its cost exceeds 1.5 times of the current optimal path, the node and its parent node are deleted.

[0038] The action performed by the controlled vehicle in the adjustment zone is the fine-tuning stage, when the rectangular neighborhood is used in the adjustment zone T5-T6, the straight path is searched reversely from the target point, and it is ensured that the angle between the vehicle body and the parking space boundary is ≤5°. The pruning rule is to remove the nodes with a distance to the obstacle less than 0.3m, a cost more than 1.2 times of the current optimal path, and an angle between the vehicle body and the parking space boundary >5°. The coarse-tuning straight path, the fine-tuning circular arc path and the fine-tuning straight path are generated by improving the hybrid A* algorithm.

[0039] Then the multi-stage path is smoothed by using piecewise B-spline interpolation, the junction T2 between the coarse-tuning and fine-tuning stages is ensured to have continuous curvature by using cubic B-spline, and the junction T5 between the fine-tuning and fine-tuning stages is reduced in calculation by using quadratic B-spline. The safety weight is dynamically adjusted according to the distance to the obstacle, when the distance to the obstacle is less than 0.5m, the safety weight is set to 2.0 to force obstacle avoidance, otherwise =1.0 to balance the smoothness and safety. The optimization objective function is solved by quadratic programming to ensure that the distance between the path and the obstacle is not less than 0.3m, and the optimization objective considers the path tracking error and the control amount increment.

[0040] Finally, the control is performed by relying on the differentiated strategy of MPC control and the retry mechanism when the obstacle intrudes. In the coarse-tuning stage, MPC uses a shorter prediction time domain Tp1 (such as Tp1=3-5 steps), focuses on minimizing the path tracking error, and the control amount penalty coefficient 1=0.5 to ensure the stability of the initial path. In the fine-tuning stage, MPC rolls the front wheel steering angle every 500ms, uses a medium prediction time domain Tp2 (such as Tp2=5-8 steps), improves the curvature adaptability, and the control amount penalty coefficient 2=1.0 to balance the steering flexibility and smoothness. In the fine-tuning stage, MPC uses an extremely short prediction time domain Tp3 (such as Tp3=2-3 steps), increases the control amount penalty coefficient 3=2.0 to limit the front wheel steering angle increment (such as Δδ≤±1°) to achieve millimeter-level position correction. When the obstacle intrusion operating zone annulus neighborhood or trapezoidal neighborhood is detected, multi-step parking retry is triggered, which can return to S3 for path planning according to the specific situation to reach the expected location T4 position, and then continue to the next step until the parking is successfully completed.

[0041] ​Through the above steps, a set of narrow parking space parking optimization method based on the intelligent driving parking area formalization design can be formed. The method constructs a parking area grid map and an environment model through multi-sensor fusion, uses rectangular neighborhood and trapezoidal neighborhood in the preparation area to complete the parking space detection and initial path planning in the coarse adjustment stage, dynamically adapts the trapezoidal neighborhood and the ring neighborhood in the operation area, and combines the improved hybrid A* algorithm of bidirectional search and heuristic pruning to realize the steering cut-in and path optimization in the fine adjustment stage. The adjustment area completes the attitude correction and accurate alignment in the fine adjustment stage by means of the fan-shaped neighborhood and the rectangular neighborhood. Combined with the path smoothing processing of segmented B-spline interpolation, the obstacle avoidance strategy of adaptive safety weight, and the synergistic effect of dynamic grid updating and MPC staged control, a full-process automatic parking solution covering from narrow parking space detection to real-time adjustment to accurate parking is formed, the smoothness, safety and real-time performance of the parking path in the narrow parking space scene are realized, the dynamic obstacle interference is effectively dealt with, and the adaptability and reliability of the automatic parking in the complex narrow environment are improved.

[0042] Although embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, replacements and changes can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain, characterized in that, Includes the following steps: S1: Construct a parking area grid map by fusing multiple sensors, detect narrow parking spaces and generate an environment model. If data is lost, parking space detection fails or type determination is incorrect, repeat this step. S2: Divide the parking area into three functional areas: preparation area, operation area and adjustment area. Define coarse adjustment, fine adjustment and micro adjustment multi-step parking stages. Each stage matches the corresponding neighborhood. If the subsequent path search has no feasible path for ≥3 consecutive times due to unreasonable area division, return to this step. S3: Perform multi-stage path search using a hybrid A* algorithm improved by bidirectional search and heuristic pruning rules. If the search results have a local optimum or no feasible path is found due to excessive pruning, adjust the parameters and re-execute this step. S4: Use piecewise B-spline interpolation to smooth the multi-step path and enhance obstacle avoidance through adaptive safety weights. If the curvature of the smoothed path changes abruptly or the safety distance is not met, return to S3. S5: Based on dynamic grid updates and MPC control, it realizes real-time adjustment of multi-step parking paths. If an obstacle enters during the parking process and causes the parking to terminate, it will return to S3.

2. The parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain according to claim 1, characterized in that: In S1, a parking space is determined to be narrow when WP2≤WP≤WP1 is satisfied, where WP is the width of the parking space, WV is the width of the vehicle body, WP1=1.35WV, WP2=1.25WV, and the narrow parking space includes vertical, parallel, and diagonal types.

3. The parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain according to claim 1, characterized in that: In S2, the vehicle's location is quickly determined by the grid index, and the parking area is divided into a preparation area, an operation area, and an adjustment area. The preparation area is used to complete the parking space detection and initial positioning as a coarse adjustment stage. This stage can use a combination of rectangular and trapezoidal neighborhoods. The operating area is the core steering entry area. Moving the vehicle into the parking space through steering and speed control is the fine-tuning stage. This stage can use a combination of trapezoidal and circular neighborhoods. The fine-tuning stage refers to the process of adjusting the vehicle after parking and finally bringing it to a complete stop. This stage can be achieved by using a combination of fan-shaped and rectangular neighborhoods.

4. The parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain as described in claim 1, characterized in that: In S3, within the region divide-and-conquer framework, the improved hybrid A* algorithm needs to be adapted to regional features to generate feasible paths, and a bidirectional multi-stage search algorithm is used to generate multi-step parking paths: Coarse adjustment phase: In the preparation area, rectangular and trapezoidal neighborhoods are used to generate a straight path from the starting point to the entrance of the operation area using an improved hybrid A* algorithm. The heuristic function is: ;in, The total cost of node n in the coarse adjustment phase. w1 represents the actual cost from the start of the coarse adjustment phase to node n, and w1 is the distance weighting coefficient. This is the Euclidean distance from the current node to the entrance of the operation area; Fine-tuning phase: Trapezoidal and circular neighborhoods are used in the operation area to initiate a bidirectional search, prioritizing the generation of circular arc paths to handle small curvature turns. The cost function is: in, The total cost of node n during the fine-tuning phase is... The actual cost from the start of the fine-tuning phase to node n. The actual cost from the target point to n. For path curvature, w1 represents the distance from the node to the obstacle, and w2 and w3 are weighting coefficients. Fine-tuning phase: Use sector and rectangular neighborhoods in the adjustment area to search for a straight path from the target point in reverse, ensuring that the angle between the vehicle body and the parking space edge is ≤5°.

5. The parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain according to claim 4, characterized in that: In S3, the heuristic pruning rules are as follows: Coarse adjustment stage: Remove nodes whose curvature exceeds the minimum turning curvature of the vehicle, and remove the path segment when the steering angle change rate of 3 consecutive nodes is >5° / s; Fine-tuning phase: Dynamic threshold pruning is adopted. When the obstacle density in the operation area is >3 obstacles / m², nodes whose cost exceeds 1.2 times that of the current optimal path are removed. When the obstacle density in the operation area is ≤3 obstacles / m², nodes whose cost exceeds 1.8 times that of the current optimal path are removed. At the same time, in bidirectional search, if a node is a leaf node (i.e., has no child nodes) and its cost exceeds 1.5 times that of the current optimal path, then the node and its parent node are deleted. Fine-tuning phase: Remove nodes that are less than 0.3m away from obstacles, cost more than 1.2 times the current optimal path, and have an angle greater than 5° between the vehicle body and the edge of the parking space.

6. The parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain according to claim 1, characterized in that: In step S4, the segmented path smoothing and safety optimization includes: using cubic B-spline interpolation for the connecting path between the coarse adjustment and fine adjustment stages to ensure curvature continuity; using quadratic B-spline interpolation for the connecting path between the fine adjustment and micro-adjustment stages to reduce computational complexity; and dynamically adjusting the path safety term based on obstacle distances. At that time, safety weight Force the path away from obstacles, otherwise Balancing path smoothness and safety; optimizing the objective function using quadratic programming (QP) to ensure the distance between the path and obstacles is no less than [value missing]. The optimization objective function is: in, This is the path tracking error vector. Q and R are weight matrices used to control the increment.

7. The parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain according to claim 1, characterized in that: In S5, the MPC control adaptation three-stage feature adopts a differentiated strategy: Coarse-tuning phase: MPC uses a shorter prediction time domain Tp1, focusing on minimizing path tracking error, and the control quantity penalty coefficient... 1 = 0.5, to ensure the stability of the initial path; Fine-tuning phase: MPC performs rolling optimization of the front wheel angle every 500ms, using a medium prediction time domain Tp2 to improve curvature adaptability and control quantity penalty coefficient. 2=1.0, balancing steering flexibility and smoothness; Fine-tuning phase: MPC uses an extremely short prediction time domain Tp3 and increases the penalty coefficient for the control input. 3=2.0, limiting the increase in front wheel steering angle to achieve millimeter-level position correction; When an obstacle is detected encroaching on the circular or trapezoidal neighborhood of the operating area, a multi-step parking retry is triggered, and the system returns to S3 to plan a path to reach the expected location, depending on the specific circumstances.

8. The parking optimization method for narrow parking spaces based on formal design of intelligent driving parking domain according to claim 7, characterized in that: In S5, the optimization objective of the MPC algorithm is: The constraint is: front wheel steering angle speed ,in For prediction error, This is the steering angle increment. This is the penalty coefficient for the control quantity.

Images